[Intermediate → Advanced] — The quantity you cannot measure, why it needs nine components, and the three different stress tensors finite deformation forces on you.
Traction
Cut an imaginary plane through a loaded body. The material on one side pulls and pushes on the material on the other. The traction is that force per unit area:
A vector, in N/m² (pascals). It has a component normal to the plane (normal stress) and components in the plane (shear stress).
The traction depends on the orientation of the cut. Same point, different plane, different answer — a horizontal cut through a bent beam sees mostly shear; a vertical cut sees mostly normal stress. That dependence is the entire reason stress is a tensor.
Cauchy’s theorem
The result that makes the field possible:
The traction on any plane is a linear function of that plane’s normal. So you don’t need infinitely many numbers to describe the stress state at a point — you need nine, and the tensor generates the traction on every possible cut.
The proof is the Cauchy tetrahedron: take a small tetrahedron with three faces on the coordinate planes and one arbitrary face with normal . Write force balance. Areas scale as and volume as , so as the body-force and inertia terms vanish faster than the surface terms — leaving a purely geometric relation. That vs argument is worth remembering; it’s used repeatedly.
Reading the components
= the force in direction on the face whose normal is .
- Diagonal — normal stresses. Positive = tension, negative = compression
- Off-diagonal — shear stresses
The sign convention matters and differs by field: solid mechanics takes tension positive; geomechanics takes compression positive, because soil is essentially always in compression and nobody wants to write minus signs all day. Check which convention a source uses.
Symmetry
This is not a definition — it’s a consequence of angular momentum balance. Take a small cube and sum moments: if , the net moment is proportional to while the rotational inertia is proportional to , so angular acceleration as the cube shrinks. An infinitesimal element would spin infinitely fast.
So symmetry is forced, and it reduces nine components to six independent ones. → Conservation Laws
(The exception: Cosserat / micropolar continua, which carry couple stresses and have non-symmetric . Relevant to granular media and materials with microstructure, and out of scope here.)
Principal stresses
Because is symmetric, it has three real eigenvalues and orthogonal eigenvectors. → note 02
In the principal directions there is no shear at all — only normal stresses .
Two facts fall out and drive most failure prediction:
Maximum normal stress is . Brittle materials — cast iron, concrete, glass, ceramics — fail on the plane perpendicular to it. Chalk twisted in torsion breaks along a 45° helix, because that’s where the principal tension lies.
Maximum shear stress is , on planes at 45° to the principal directions. Ductile materials yield by shear, so mild steel in torsion fails on a transverse plane. Same test, different fracture surface, and the orientation tells you the failure mode. → Failure and Yield
Mohr’s circle is the graphical construction of this in 2D — a circle of centre and radius , on which every plane’s stress state is a point. It’s how this was done before computers, it’s still the fastest way to build intuition, and it’s still on every exam.
The volumetric/deviatoric split
Pressure changes volume; the deviator changes shape.
This split does real work:
- Metals yield on the deviator only. Hydrostatic pressure doesn’t cause plastic flow — a solid steel ball at the bottom of the Mariana Trench is fine, a hollow one implodes. That’s why the von Mises criterion is built from
- In fluids, is the thermodynamic pressure and is the viscous stress. A fluid at rest has and — pure pressure, no shear, which is the definition of a fluid at rest. → Navier–Stokes
- Rubber is nearly incompressible, so the hydrostatic part is essentially unconstrained by deformation and must be carried as a separate unknown — which is why hyperelastic FE formulations use mixed elements
Three stress tensors
Under finite deformation the question “force per unit area” becomes ambiguous: which area — deformed or undeformed? That question generates three tensors.
Cauchy stress — “true stress.” Current force over current area. The physically real one; what a material actually experiences and what failure criteria use. Symmetric. Spatial.
First Piola–Kirchhoff — “nominal / engineering stress.” Current force over reference area:
Convenient because the reference area is known, and it’s what a tensile machine effectively reports (load ÷ original cross-section). Not symmetric, being a two-point tensor like — which makes it awkward for constitutive laws.
Second Piola–Kirchhoff . Both force and area pulled back to the reference configuration:
Symmetric, fully material, and energy-conjugate to Green–Lagrange strain:
It has no direct physical interpretation — it’s the force pulled back through , which isn’t a force you could measure. It exists because the mathematics is clean, and that’s enough: hyperelastic constitutive laws are written as , and FE codes work internally in and , converting to only for output.
The practical rule: use Cauchy stress for anything physical — failure, reporting, comparison with experiment. Use and inside a finite-deformation formulation because the pair is symmetric and conjugate. In small strain all three coincide, which is why undergraduate courses never mention it.
The engineering-vs-true stress distinction on a tensile curve is exactly vs : the engineering curve falls after the UTS because it keeps dividing by the original area while the specimen necks; the true curve keeps rising.
You cannot measure stress
Worth stating plainly, because it’s easy to forget.
There is no stress gauge. What you measure is force (a load cell) or strain (a gauge or DIC), and stress is inferred:
So every stress number carries the assumptions of the model used to compute it. If the material yielded and you used a linear-elastic relation, your “measured” stress is wrong — and it will look entirely plausible.
Photoelasticity and X-ray diffraction come closest to direct measurement, and both still infer through a physical model.
This is a good instance of the general point from the domain note: the balance laws are certain, the constitutive model is chosen, and that’s where the error lives.
Equilibrium
Static force balance, which is what most of solid mechanics solves:
The divergence of the stress tensor plus body force per unit volume equals zero. Three equations, six unknown stress components — underdetermined, which is why you need compatibility and a constitutive law to close the system.
With acceleration it’s Cauchy’s equation of motion, and it’s the same equation that becomes Navier–Stokes for a fluid and the Navier equations for an elastic solid. One equation, two fields. → Conservation Laws
Related
- Strain Measures — the conjugate quantity
- Conservation Laws — where equilibrium comes from
- Failure and Yield — what principal stresses predict
- Continuum mechanics map
Software engineering
Infrastructure & security
Computer science
Hardware & engineering